support for creating ptys and a login_tty fork_action

[avsm]

Just a draft to figure out the right way to support spawning a shell with a pty:

• [ ] do the pty fds need to created as cloexec, and then inherited in a fork_action to the subshell? Right now they are not cloexec, but there is no close Fork_action and so there is too much open on the other side.
• [x] we use login_tty as a compromise between manually doing the tty setup ourselves, and do not need low-level bindings to setsid and other terminal manipulation bits.
• [ ] unsure how to expose this more portably in Eio, as right now Fork_actions require a direct dependency on either Eio_linux or Eio_posix, and cannot be called via Eio_unix. See https://github.com/avsm/eeww/commit/3cbac4c5142727f0cd28f6e0babb57b4b5e3c3e4#diff-81dfb4951962ecf602592af852995decc04ccce063dd78977949ed086c803adfR45 for some code that mimics Eio.Process.spawn, but with a pty. Should this plumbed through Eio.Process.spawn somehow?
• [ ] still need to add tests
• [ ] only Linux at the moment, need to do the configure bits to work on macOS/*BSD

/cc @RyanGibb

[RyanGibb]

This looks great! Using login_tty is much nicer than manually doing tty setup.

[ ] do the pty fds need to created as cloexec, and then inherited in a fork_action to the subshell? Right now they are not cloexec, but there is no close Fork_action and so there is too much open on the other side.

I saw that they're manually closed in https://github.com/avsm/melange/blob/e92240e6dc8a440cafa91488a1fc367e2ba57de1/lib/ounix/ounix.ml#L66. Perhaps they should similarly be closed in a Pty.close after the fork returns? I think setting cloexec would require manually creating a pty by opening /dev/ptmx (/dev/pty on bsd I think), and calling ioctl TIOCGPTPEER, etc, as openpty doesn't doesn't support it.